This is a repository for the first two problems of my Master's Thesis called Coarse-Graining of Molecular Dynamics Using Neural Ordinary Differential Equations. The project has two parts. Firstly, three different toy potentials are learned with neural ODEs. Then a benchmark is done by performing Hamiltonian Monte Carlo sampling of those potentials, where the dynamics proposal step is evolved with neural ODEs. Secondly, a diatomic harmonic molecule potential is learned and then extended to a triatomic molecule, investigating the feasibility of extending pair-wise interactions.
This project was done as a collaboration between MIT's Rafael Gómez-Bombarelli and Imperial's Stefano Angiolleti-Uberti research groups.